scCausalVI disentangles single-cell perturbation responses with causality-aware generative model.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cellular heterogeneity.
• Distinguishing intrinsic cellular variation from external stimuli effects is challenging.
• Accurate deconvolution is crucial for understanding cellular responses.

Purpose of the Study:

• Introduce scCausalVI, a causality-aware generative model.
• Disentangle intrinsic cellular states from external perturbation effects.
• Improve analysis of scRNA-seq data for biological insights.

Main Methods:

• Developed a deep structural causal network to model causal mechanisms.
• Integrated structural causal modeling with in silico prediction.
• Accounted for technical variations and cell-state-specific responses.

Main Results:

• scCausalVI effectively disentangles causal relationships and quantifies treatment effects.
• The model generalizes to unseen cell types and separates biological from technical variation.
• Applied to COVID-19 data, it identified treatment-responsive populations and susceptibility signatures.

Conclusions:

• scCausalVI offers a robust framework for causal inference in scRNA-seq data.
• The model enhances the ability to interpret cellular responses to perturbations.
• Provides a powerful tool for analyzing complex biological systems and disease mechanisms.